Research On Personalized Recommendation Based On Collaborative Filtering

With the development of E-commerce, personalized recommendation, as one ofthe key techniques has been widespread concern. Based on the user’s personalcharacteristics, historical behavior and the feature of goods or other information,personalized recommendation uses collaborative filtering, content filtering,knowledge discovery, interactive technology or other recommended technology torecommend their goods to the user which the user may be interested in.Among diverse recommendation algorithm, collaborative filteringrecommendation algorithm is currently one of the most widely used technologies.However, when applied to e-commerce personalized recommendation, collaborativefiltering algorithm face the following issues: firstly, the problem how to consider theuser’s interest changes over time can make the calculations of similarity between theusers more precise; secondly, how to use social networks to more accurately measurethe similarity between users can make the nearest neighbor of users be more accurate;and thirdly, the problem how to consider the behavior of users which have the sameinterests have similarity score not the same score can make the predicted score ofgoods more accurate, fourthly, how to use the inherent relationship between projectcategories, such as internal relations and orderly relations can provide users with moreaccurate recommendation results.To solve these problems, this paper integrates timing updates, trust, optimizationof predicted score and structured thought with the traditional collaborative filteringalgorithm. First of all, improved algorithm uses the idea of the timing update todistinguish the user’s long-term interests and recent interest, so that the algorithm canaccurately calculate the current similarity between users. Then when taking advantageof the user’s social network, this algorithm adds trust to similarity as a deeperexplanation and finds the nearest neighbor. Next the improved algorithm takesadvantage of the goods’ score evaluated by this nearest neighbor to predict the user’sscore on goods by optimized the score predicted strategies, and recommends thesegoods which have higher predicted score to the user. Finally this algorithm takes theinherent relationship between goods categories into account to find category that therecommended goods is in, and then locks the category of secondary recommended items, and considers the characteristics of users and goods recommended to provideusers with a secondary recommendation.Finally, experiment on MovieLens datasets is done to compare therecommendation results between the improved personalized recommendation and thetraditional method of collaborative filtering recommendation algorithm and resultsshow that the improved personalized recommendation has a higher accuracy.